406 



liave beêti propagated from tlie "Stermeter" and have badly influenced 

 the optical observations. 



The only purpose however was to compare the indications of the 

 two meters. By two independent, quite corresponding measurements 

 on different days it was proved, that the large meter gave 10000 L., 

 when the accurate "Ster" meter registrated 9810 L. only. This is 

 a difference of 1.97o- Now tiie error of the "Ster" meter itself 

 is about O.27o as iiad been determined liy direct measurement 

 of the volume transmitted to a large tank on the grounds of the 

 waterworks. Altogether, taking the error of 0.27» with the right 

 sign, the error in the indication of the large watermeter amounts 

 to 2.1 7u- We have seen already, that theory and experiment 

 agree extremely well, if we introduce this correction, which I'ednces 

 the values of ?«„ from 475 c.m./sec. to 465 c.ra./sec. for p = 2.14 

 k-g./cm'-*. 



The value of w„ at p = 2.14 K.g./c.m." may thus be regarded as 

 well established and the same may be said of the value of /, at least 

 withiji the limits of the accuracy of the final result. About the 

 factor 0.84 however some doubt may exist. Therefore it seems to 

 us interesting to show, that even if the (z/Wu/t? value of the Lorentz 

 dispersion-term might have been determined less accurately than has 

 been the case, there might have been drawn a conclusion about (he 

 necessary existence of this term '). This conclusion is independent of 

 the values given to I and //'„,„.r. . 



For, writing down ecpiation (4) for two difïerent colours with the 

 wavelengths A, and X^ we see, that /, iVmax. and c fall out by the 

 division. The ratio of the shifts A^.^ and ZiA, becomes then according 



to LORENTZ 



(Ó) 



and according to Fresnel 



1) I will slill make one remark. If we wished to explain the difference of 5 % 

 between our observations and the formula of Fresnel by an error in the factor 

 0,84, we should have to change tliis factor into 0,88 in order to obtain coin- 

 cidence of the experimental curve and that of Fresnel. But such a great inaccuracy 

 does by no means exist in that factor. 



